Rice canopy biochemical concentration retrievals based on Hyperion data

Transcription

1 (2009) Journal of Remote Sensing 遥感学报 Rice canopy biochemical concentration retrievals based on Hyperion data CHEN Jun-ying 1,2, TIAN Qing-jiu 1, QI Xue-yong 2, LIU Xiao-chen 1, GUAN Zhong 1 1. International Institute for Earth System Science, Nanjing University, Jiangsu Nanjing , China; 2. China Centre for Resources Satellite Data & Application, Beijing , China Abstract: The method of plot experiment with field application was used in this research. The variations of rice foliage and canopy spectra with corresponding biochemical concentration field-measured in jointing, heading and filling stages were analyzed. Through the analysis of absorption characteristic and vegetation index, the best spectral features for rice nitrogen and chlorophyll concentration estimation were obtained. Based on the Hyperion image of Jiangyan, we built the models of rice canopy nitrogen and chlorophyll estimation. At last, rice canopy nitrogen and chlorophyll concentration were retrieved from Hyperion image and their distribution maps were obtained. The results showed that: (1) Nitrogen concentration can be retrieved accurately using the area of absorption feature centered at 670nm based on band depth normalized to band depth (BNC) analysis; (2) Based on reversional normalized spectrum, NDVI using 560nm and 670nm was strongly correlated with chlorophyll concentration. Key words: hyperspectral remote sensing, rice, Hyperion, nitrogen, chlorophyll CLC number: TP79 Document code: A 1 INTRODUCTION Precision agriculture is the world trend of agricultural development, and also an important way for agricultural sustainable development (Liu, 2002). Through the application of 3S technology, the space and time difference information of environmental factors related with crop growth can be collected and processed, and the management can be accurately adjusted, and then the maximum yield and economic efficiency can be obtained without environmental damage (Zhu et al., 2000). Rice is one of the main food crops in China. Biochemical parameters are good indicators of rice growth status. For example, nitrogen concentration is the important guiding factor of rice fertilization, while chlorophyll concentration can reflect the situation of rice photosynthesis. Real-time monitoring of changes in rice canopy biochemical concentration can help us to understand the growth of large area rice in time, and it is an important aspect of precision agriculture (Li, 2003). Hyperspectral remote sensing has the characters of high spectral resolution and good continuity of bands, which can provide more spectral information. The development of hyperspectral remote sensing provides a powerful tool for quantitatively analyzing the relationship of biochemical concentration and spectral features. At present there are many methods for biochemical concentration retrievals, but most of them are based on the field measured spectra. Even if the remote sensing images were used, most of them are aerial images such as Hy- Map, AVIRIS, OMIS and so on. There are not many articles describing the rice biochemical concentration quantitative retrieval and mapping based on hyperspectral satellite imagery. This study was based on spaceborne hyperspectral remote sensing technology and take precision agriculture as the application target. Rice foliage, canopy spectra and biochemical concentration (nitrogen & chlorophyll) of different species, growth conditions and at different growth stages were analyzed to research on the correlation model between spectral features and biochemical concentration. Furthermore, quantitative extraction and mapping technologies of rice canopy biochemical concentration based on hyperspectral remote sensing information were studied. 2 DATA SOURCES 2.1 Research area The method of plot experiment with field application was used in this research. The study site is located at Yangzhou University ( E, N), in Jiangsu province, China. The total area is 715m 2. Yangzhou is located in the centre of Jiangsu province, the north shore of the Yangtze River and the southern end of the Jianghuai plain. Yangzhou has high index of land reclamation. The soil fertility and productivity of Yangzhou are relatively high in Jiangsu province. Rice planted Received: ; Accepted: Foundation: Jiangsu Povince High Tech Research Projects (No.BG and No. BG ). First author biography: CHEN Jun-ying(1983 ), female, master. She graduated form Nanjing University, and is engaged in research of hyperspectral remote sensing. She has published 5 papers. cjy831025@163.com.

2 CHEN Jun-ying et al.: Rice canopy biochemical concentration retrievals based on Hyperion data 1107 in this field included two typical varieties: Changyou 1 and Wujing 15. Nitrogen application from 0 to 25 kg was divided into 8 levels. There are two same series and 32 plots. The field covered by Hyperion image for application was located in Jiangyan area, Jiangsu province. Jiangyan has been called granary since ancient times. Jiangya teemed with rice and it is the commodity grain base of China. We selected eight paddy fields of image covering area as sampling sites. Synchronously with the image acquisition, field-based canopy reflectance spectra, nitrogen and chlorophyll concentration were recorded at those eight sites. The latitudes and longitudes of those sites were accurately measured by GPS, in order to precisely locate the sites in the image. Fig. 1 shows the image coverage and sampling sites distribution. The foliage and canopy reflectance spectra were collected with ASD FieldSpec Pro FR spectroradiometer. The wavelength range is from 350 to 2500 nm. Sampling interval between 350 and 1000 nm is 1.4 nm, between 1000 and 2500 nm is 2 nm. Rice canopy reflectance spectra were collected when the sky was clear. Probe perpendicular to the canopy, and the distance between probe and rice canopy was about 0.5 m, 25 visual field was used. We selected 3 5 sample points in every plot and collected 2 3 spectra for every sample point. All of the spectra collected in one plot were averaged to obtain the canopy spectrum of that plot. We collected 5 rice leaves of every plot in Yangzhou experimental field, 10 rice leaves of every plot in Jiangyan field for foliage spectra measurement. The foliage spectra were collected in darkroom. The reflectance spectrum of every leaf was acquired by averaging the spectra of 4 6 measurements. This study was from August 15 to September 30, 2005, involved jointing, heading and filling stages canopy spectra and 2925 foliage spectra were collected. As shown in Table 1, we measured the canopy and foliage reflectance spectra in Yangzhou experimental field on August 15, September 6 and September 30, Synchronously with the image acquisition, field-based canopy reflectance spectra were recorded at 8 sample plots in Jiangyan field on September 7, Table 1 Spectra field-measurement in 2005 Date August 15 September 6 September 7 September 30 Location Growth stage Sample number Canopy spectra number Foliage spectra number Yangzhou experimental field Jointing stage Yangzhou experimental field Heading stage Jiangyan field Heading stage Yangzhou experimental field Filling stage Nitrogen and chlorophyll concentration measurement and treatment Fig. 1 The image coverage and sampling sites distribution 2.2 Field spectra measurement Nitrogen and chlorophyll concentration was immediately measured after the spectra acquisition. Nitrogen concentration was measured through kjeldahl nitrogen method. Chlorophyll concentration was measured by using the SPAD-502 instrument. The SPAD-502 instrument can measure foliage chlorophyll concentration without destruction. It determines foliage chlorophyll relative concentration according to the absorptivity of two wavelength ranges (Pu & Gong, 2000). The chlorophyll concentration was acquired by averaging the values of 6 to 10 measurements. The nitrogen and chlorophyll concentration of every plot was determined by averaging all the values of leaves sampled from the plot. 2.4 Hyperion data acquisition The Hyperion sensor is the first civilian spaceborne hyperspectral instrument. It acquire visible near-infrared (VNIR) and shortwave infrared (SWIR) spectra by push broom technique. Hyperion data have 242 bands and the main characteristics are shown in Table 2. The Hyperion data was acquired at 10:20 A.M. on Sep 7, 2005, with a size of pixels. The image covered Jiangyan area, Jiangsu province and was centered at E, N. Table 2 Main characteristics of Hyperion data (Richard, 2003) Wavelength range/nm Number of bands 242 Spatial resolution/m 30 VNIR bands 1 70 ( nm) SWIR bands ( nm) Data type 2-byte singed integer

3 1108 Journal of Remote Sensing 遥感学报 2009, 13(6) 3 HYPERION IMAGE PROCESSING 3.1 Hyperion image preprocessing Hyperion image preprocessing included eliminating uncalibrated and atmospheric water vapor affected bands, absolute radiometric calibration, stripes and smile effect removal, atmospheric correction and spectrum smoothing. At last, the reflectance image was obtained. A number of bands were uncalibrated and others had low sensitivity of the spectrometer materials. Because of this, only 166 bands were used in this study. The remain bands and wavelengths are shown in Table 3. Table 3 The remain wavelengths of Hyperion image Hyperion original bands Wavelengths/nm The SWIR bands have a scaling factor of 80 and the VNIR bands have a scaling factor of 40 applied (Richard, 2003). Every band dividing by the scaling factor can generate the absolute radiometric calibrated image. Global equalization was used to remove the stripes and smile effect. It thought that the mean value and standard deviation of each column in each band were the same. The mean value and standard deviation of each column were modified to match those for the whole image for each band. The image was corrected for atmosphere with the Fast Line-of-sight Atmospheric Analysis of Spectral Hypercubes (FLAASH) to retrieve the ground reflectances. FLAASH is a first-principles atmospheric correction modeling tool for retrieving spectral reflectance from hyperspectral images. It made the obscuring effects of the atmosphere accurately removed. After a series of processing, the deficiencies and uncertainties of every step lead to the continuous noise of atmospheric corrected image spectra. The noise may exert an adverse effect on the quantitative application of image. An inverse minimum noise fraction (MNF) transformation was applied to the FLAASH-derived surface reflectance image to separate noise from the data (Datt et al., 2003). The atmospheric corrected image reflectances were quite similar to the field measured canopy reflectances (Fig. 2). The result of spectral rebuilt was satisfactory. 3.2 Geometric correction The geometric correction was through applying 22 GCPs collected from 1:50000 relief maps covering the study area to the reflectance image. The total error was less than one pixel. 3.3 Discrimination of paddy rice fields The research subject of this paper is rice, so we discrimi- Fig. 2 Comparison between Hyperion reflectance spectrum and field measured canopy spectrum nated paddy rice field from the image for rice biochemical concentration analysis and retrieval. It was also help for identifying the results more easily. Firstly, vegetation was extracted with normalized difference vegetation index (NDVI). NDVI is given by the formula: ρnir ρ NDVI = RED (1) ρnir + ρred where ρ RED and ρ NIR are the reflectance of red and near infrared bands. We applied the threshold (NDVI 0.7) to identify vegetation pixels. Then we calculated Enhanced Vegetation Index (EVI) and Surface Water Index (LSWI) of the image (Huete et al., 2002). EVI and LSWI are given by the formulas: ρ EVI 2.5 NIR ρ = RED (2) ρ + 6 ρ 7.5 ρ + 1 NIR RED BLUE ρnir ρ LSWI = SWIR ρnir + ρswir where ρ BLUE, ρ RED, ρ NIR and ρ SWIR are the reflectance of blue, red, near infrared and shortwave infrared bands. The Hyperion image was acquired at the heading stage of rice and the leaf area index was relatively higher at that time. As we know when the vegetation coverage more than 80%, the NDVI values increase with delay and become saturation, so in this study we chose Enhanced Vegetation Index (EVI) to identify rice from other vegetation. EVI directly adjusts the reflectance in the red band as a function of the reflectance in the blue band, and it accounts for residual atmospheric contamination and variable soil and canopy background reflectance (Huete et al., 1997, 2002). The EVI values of paddy rice fields were lower than other vegetation fields. The distinct differentiate between paddy rice fields and other vegetation fields is that the rice is grown on flooded soils, so we used LSWI which was sensitive to the leaf water and soil moisture. The LSWI values of paddy rice fields were higher than other vegetation fields. Thus it can be seen as to paddy rice fields, the values of EVI LSWI were lower than other vegetation fields. In this study, the paddy rice (3)

4 CHEN Jun-ying et al.: Rice canopy biochemical concentration retrievals based on Hyperion data 1109 field pixels were identified by using the threshold of EVI LSWI Fig. 3 shows the comparison between original pseudo-color composite image and paddy rice image. As the original pseudo-color composite image of Fig. 3(a) shows, paddy rice fields are presented as dark red and other vegetation fields are presented as bright red. Paddy rice fields are presented as white in Fig. 3(b). The accuracy of paddy rice field discrimination was evaluated with 3812 non-vegetation samples, 272 rice samples and 2483 other vegetation samples. We established confusion matrix and computed some correlative data. The accuracy of paddy rice field discrimination is shown in Table 4. The overall accuracy of paddy rice field discrimination is 92.89% and the kappa is The result showed that the method of paddy rice field discrimination was relatively accuracy and feasible. Fig. 3 Comparison between original pseudo-color composite image and paddy rice image (a) Original pseudo-color composite image; (b) Paddy rice image Non-vegetation Type Non-vegetation Table 4 Rice discrimination accuracy Rice Other vegetation Sum User accuracy/% Rice Other vegetation Sum Producer accuracy/% Overall accuracy=92.89% Kappa coefficient= The method was based on the thought of continuum removal can remove those background influence and thus to isolate individual absorption features. The continuum is a straight line fitted over the top of a spectrum that connects local spectral maxima. The continuum-removed reflectance is calculated by dividing the original reflectance values by the reflectance of continuum at the corresponding wavelength (Shi et al., 2005). Based on the continuum removal, the spectral transformation of band depth normalized to band depth at the centre of the absorption feature (BNC) and the area of absorption feature (BNA) were used Band depth normalized to band depth at the centre of the absorption feature (BNC) The normalized band depths (BNC) are calculated by dividing the band depth of each band by the band-depth at the band center: BNC = (1 ( R/ R')) /(1 ( R / R )) (4) where R is the reflectance of each wavelength, R is the reflectance of continuum at the corresponding wavelength, R c is the reflectance of the absorption feature centre wavelength, R c is the reflectance of continuum at the absorption feature centre wavelength Band depth normalized to area of absorption feature (BNA) The band depths normalized to area (BNA) are calculated by dividing the band depth of each band by the area of the absorption feature: BNA = (1 ( R / R')) / A (5) where A is the area of the continuum-removed absorption feature Selection of spectral absorption feature band In this study, we used two band selection methods: (1) We applied continuum-removal analysis to the whole spectrum of each sample. As shown in Fig. 4 and Fig. 5, the continuum-removal spectrum produced four isolated wavelength ranges centered at 670, 980, 1160 and 1950nm. (2) We selected 6 wavelength ranges according to the rice original reflectance spectrum and the known chlorophyll and nitrogen absorption feature (Table 5). We consulted the result of absorption features listed by Curran (1989). The nitrogen absorption feature bands are 910, 1020, 1510, 1980, 2060, 2130, 2180, 2240, 2300 and 2350nm, the chlorophyll absorption feature bands are 460, 640 and 660nm. In this study, we analyzed the correlation of nitrogen and chlorophyll concentration with the depth, width and area of BNC and BNA absorption feature. c c 4 METHODS 4.1 Method based on spectral absorption feature Fig. 4 A reflectance spectrum and its continuum

5 1110 Journal of Remote Sensing 遥感学报 2009, 13(6) We consulted some literatures and selected 6 vegetation indices shown in Table 6 according to the rice spectral features. We used the spectral transformation of 1 minus continuum-removed spectra and reversional normalized spectra were obtained. We calculated the 6 vegetation indices based on original spectra, normalized spectra and reversional normalized spectra. The correlation coefficients between biochemical concentration (nitrogen and chlorophyll) and vegetation indices can be obtained. Fig. 5 A continuum-removed spectrum and the band centers 4.2 Method based on vegetation indices Table 5 Selected wavelength ranges and their associated absorption features for nitrogen and chlorophyll Known chlorophyll and nitrogen Selected wavelength/nm absorption features/nm , , , 2060, 2130, , 2300, Table 6 Vegetation indices employed in this study Vegetation index Full name Formula Reference NDVI560_670 Normalized difference vegetation index RVI560_670 Ratio vegetation index NRI Nitrogen reflectance index CARI Chlorophyll absorption ratio index VARI_green Visible atmospherically resistant index VARI_700 Visible atmospherically resistant index R560 R670 NDVI560_ 670 = R560 + R670 Rouse et al., 1974 R560 RVI560 _ 670 = Pearson & Miller, 1972 R670 R570 R670 NRI = R570 + R670 a R670 + b R700 CARI = 2 a + 1 R670 a= ( R700 R550)/150, b= R a R560 R670 VARI_green = R560 + R670 + R450 R R R450 VARI _ 700 = R R R450 Schleiche et al., 2001 Kim et al., 1994 Anatoly et al., 2002 Anatoly et al., RESULT ANALYSIS 5.1 Selection of best spectral feature for rice nitrogen and chlorophyll concentration estimation We analyzed the correlation between field spectra measured on August 15, September 6 and September 30 with biochemical concentration. The correlation coefficients between biochemical concentration with spectral features and vegetation indices were calculated. Table 7 shows the spectral features significantly correlated with nitrogen concentration in the 3 growth stages. The spectral feature with maximum correlation coefficient is 670Area (BNC), which means the area of absorption feature centered at 670nm based on BNC analysis. So the 670Area (BNC) was chosen to be the best spectral feature for nitrogen concentration retrieval. Table 8 shows the spectral features significantly correlated with chlorophyll concentration in the 3 growth stages. The spectral feature with maximum correlation coefficient is NDVI560_670 (reversional normalized spectrum), which means NDVI used 560nm and 670nm based on reversional normalized spectrum. So the NDVI560_670 (reversional normalized spectrum) was chosen to be the best spectral feature for chlorophyll concentration retrieval. 5.2 Rice canopy nitrogen and chlorophyll concentration retrievals and mapping based on Hyperion data The above best spectral features for nitrogen and chlorophyll concentration retrievals based on field measured spectra study were applied to the Hyperion image reflectance of 8 fields. The rice canopy nitrogen and chlorophyll concentration retrieval models are: N = Area(BNC) (6) CHL= NDVI560_670(reversional normalized spectrum) (7) The quadratic polynomial multiple correlation coefficient (R 2 ) between the 670Area (BNC) and nitrogen concentration is 0.79 (Fig. 6). The R 2 between the NDVI560_670 (reversional normalized spectrum) and chlorophyll concentration is 0.77 (Fig. 7).

6 CHEN Jun-ying et al.: Rice canopy biochemical concentration retrievals based on Hyperion data 1111 Table 7 Spectral features significantly correlated with nitrogen concentration and the correlation coefficients Spectral features August 15 September 6 September 30 Foliage Canopy Foliage Canopy Foliage Canopy PSSRb (original spectrum) PSNDb (original spectrum) CARI (normalized spectrum) NDVI560_670 (reversional normalized spectrum) RVI560_670 (reversional normalized spectrum) NRI (reversional normalized spectrum) CARI (reversional normalized spectrum) VARI_green (reversional normalized spectrum) VARI_700 (reversional normalized spectrum) Area (BNC) Depth (BNA) , 660Depth (BNA) Table 8 Spectral features significantly correlated with chlorophyll concentration and the correlation coefficients August 15 September 6 September 30 Spectral features Foliage Canopy Foliage Canopy Foliage Canopy PSNDb (original spectrum) CARI (original spectrum) NDVI560_670 (reversional normalized spectrum) RVI560_670 (reversional normalized spectrum) NRI (reversional normalized spectrum) CARI (reversional normalized spectrum) VARI_green (reversional normalized spectrum) VARI_700 (reversional normalized spectrum) Depth (BNA) , 660Area (BNC) , 660Depth (BNA) Fig. 6 Correlation between 670Area (BNC) based on image spectrum and nitrogen concentration We applied the two models on the Hyperion paddy rice image and produced the chlorophyll and nitrogen concentration distribution maps (Fig. 8, Fig. 9). The values of nitrogen Fig. 7 Correlation between NDVI560_670 (1-Normalized spectrum) based on image spectrum and rice canopy chlorophyll concentration concentration in nitrogen concentration distribution map ranged from 0.06 % to 4.46 % and the values of chlorophyll concentration in chlorophyll concentration distribution map ranged from

7 1112 Journal of Remote Sensing 遥感学报 2009, 13(6) Fig. 8 Rice canopy nitrogen concentration distribution map Fig. 9 Rice canopy chlorophyll concentration distribution map to They were quite consistent with those of field measurements. The pixel values in each field were relatively uniform. The distribution agreed highly with the growth status distribution. Therefore to some extent, the estimation equations were validated. 6 CONCLUSIONS In this study, we analyzed the correlation between rice biochemical concentration (nitrogen and chlorophyll) with field measured foliage and canopy spectra of different species and different growth stages. Then we obtained the universal spectral features with high precision for rice nitrogen and chlorophyll concentration estimation based on field measured spectra analysis. The result of field measured spectra analysis was applied to Hyperion image and the nitrogen and chlorophyll concentration estimation models were established. The study investigated the feasible of nitrogen and chlorophyll concentration retrieval based on Hyperion image and retrieved the rice canopy nitrogen and chlorophyll concentration of research area. At last, the rice canopy nitrogen and chlorophyll concentration distribution maps were produced. The results include as follow: (1) The area of absorption feature centered at 670nm based on BNC analysis was found to be strongly correlated with the nitrogen concentration of jointing, heading and filling stages both at foliage and canopy level. The rice canopy nitrogen concentration retrieval model was built based on spaceborne hyperspectral remote sensing image spectra using the spectral feature. The results showed that the correlation coefficient (R 2 ) between the area of absorption feature centered at 670nm based on BNC spectra and rice canopy nitrogen concentration was The result of rice canopy nitrogen concentration retrieval was quite accurate and reliable. (2) Based on the reversional normalized spectrum, 560nm and 670nm were used to establish NDVI560_670. The spectral feature was found to be strongly correlated with the chlorophyll concentration of jointing, heading and filling stages both at foliage and canopy level. The rice canopy chlorophyll concentration retrieval model was built based on spaceborne hyperspectral remote sensing image spectra using the spectral feature. The results showed that the R 2 between NDVI560_670 based on the reversional normalized spectra and rice canopy chlorophyll concentration was The result was satisfactory. The study taken rice as the research object, investigated the methods for nitrogen and chlorophyll concentration retrievals. Other vegetation types and biochemical parameters can be taken as the research objects in future. The results of this study can provide reference for estimation of crop yield and pests detection with hyperspectral remote sensing. REFERENCES Anatoly A, Gitelson Y, Yoram J K, Robert S and Don R Novel algorithms for remote estimation of vegetation fraction. Remote Sens. Environ, 80: Curran P J Remote sensing of foliar chemistry. Remote Sensing of Environment, 30, Datt B, McVicar T R, Van Niel T G, Jupp D L B and Pearlman J S Pre-processing EO-1 hyperion hyperspectral data to support the application of agricultural indices. IEEE Transactions on Geoscience and Remote Sensing, 41(6): FLAASH Module User s Guide, ENVI FL ASH Version 4.2, August, 2005 Edtion <

8 CHEN Jun-ying et al.: Rice canopy biochemical concentration retrievals based on Hyperion data 1113 < Huete A R, Liu H Q, Batchily K and VanLeeuwen W A comparison of vegetation indices global set of TM images for EOSMODIS. Remote Sensing of Environment, 59: Huete A R, Didan K, Miura T, Rodriguez E P, Gao X and Ferreira L G Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sensing of Environment, 83: Kim M S The Use of Narrow Spectral Bands for Improving Remote Sensing Estimation of Fractionally Absorbed Photosynthetically Active Radiation (fapar). Masters Thesis, Department of Geography, University of Maryland, College Park Li Y M, Ni S X and Wang X Z The robustness of linear regression model in rice leaf chlorophyll concentration prediction. Journal of Remote Sensing, 9(5): Liu L Y Applications of Hyperspectral Remote Sensing In Precision Agriculture. Postdoctoral research work report. Dissertation of Institute of Remote Sensing Application Pearson R L and Miller L D Remote mapping of standing crop biomass for estimation of the productivity of productivity of the short-grass prairie. Pawnee National Grasslands, Colorado. Proceedings of the 8 th International Symposium on Remote Sens. Environ., ERIM International. Pu R L and Gong P Hyperspectal Remote Sensing and The Application. High Education Press, Beijing. Richard B. 2003, EO-1 User Guide v. 2. 3, and Rouse J W, Haas R H, Schell J A, Deering D W and Harlan J C Monitoring the Vernal Advancement of Retrogradation of Natural Vegetation. NASA/GSFC, Type III, Final Report, Greenbelt, MD, USA Schleicher T D, Bausch W C, Delgado J A and Ayers P D Evaluation and refinement of the nitrogen reflectance index (NRI) for site-specific fertilizer management ASAE Annual International Meeting. St. Joseph, MI, USA Shi R H, Niu Z and Zhuang D F Research on the effects of leaf biochemical concentrations on leaf spectra: case study of inversion of C:N ratio based on the absorption features centered at 2100nm. Journal of Remote Sensing, 9(1): 1 7 Zhu S P, Chen J, Chen Y Z, He P X and Li Y W The prototype research of precision agriculture computer integrated manufacturing system. Transactions of the CSAE, 16(4):

9 1114 Journal of Remote Sensing 遥感学报 2009, 13(6) Hyperion 陈君颖 1,2, 田庆久 1, 亓雪勇 2, 刘晓臣 1, 管仲 1 1., ; 2., 摘要 :,,, ; Hyperion,, :, 670nm ;, 560nm 670nm, NDVI560_670 关键词 :,, Hyperion,, 中图分类号 : TP79 文献标识码 : A 1, (, 2002) 3S,,,, (, 2000),,,,, (, 2003),,,,, HyMap AVIRIS OMIS,, ( ),, ( E, N), 715m 2,,,,, 1 15, 25kg 8,, 32 Hyperion,,, 8,, GPS, 1 收稿日期 : ; 修订日期 : 基金项目 : ( : BG BG ) 第一作者简介 : (1983 ),,,, 5 cjy831025@163.com

10 : Hyperion , ASD FieldSpec Pro FR, nm, nm 1.4nm, nm 2nm, 25, 0.5m, 3 5, 2 3, 5, 10, 4 6, , 3, , 表 年地面光谱采集情况 / / / SPAD-502 SPAD-502, (, 2000) 6 10, 2.4 Hyperion Hyperion, (VNIR) (SWIR), 242, Hyperion E, N, 表 2 Hyperion 数据主要技术参数 (Richard, 2003) /nm /m 30 VNIR 1 70( nm) SWIR ( nm) 2 3 Hyperion 3.1 Hyperion Hyperion Smile, Hyperion, 166, 3 VNIR 40,, SWIR 80,

11 1116 Journal of Remote Sensing 遥感学报 2009, 13(6) 表 3 保留的 Hyperion 影像波段 Hyperion /nm ,, Smile Hyperion, Fast Line-of-sight Atmospheric Analysis of Spectral Hypercubes(FLAASH) Hyperion, FLAASH,,,,,, MNF (Datt, 2003) ( 2), 2 Hyperion ( ) ( ) , 22, Hyperion, 3.3,, (NDVI), : ρ NDVI = ρ NIR NIR ρ + ρ RED RED (1), ρ RED ρ NIR NDVI 0.7 (EVI) (LSWI)(Huete, 2002) : ρnir ρred EVI = 2.5 ρ + 6 ρ 7.5 ρ + 1 NIR RED BLUE ρ LSWI = ρ NIR NIR ρ + ρ SWIR SWIR (2) (3), ρ BLUE ρ RED ρ NIR ρ SWIR 9 7,,, 80%, NDVI,,, EVI,, (Huete, 1997, 2002), EVI, LSWI, LSWI, EVI LSWI EVI LSWI (a),, 3(b), ,,, ( 4) 4, 92.89%, Kappa ,

12 : Hyperion (a) ; (b) : (1) 4 5, 4, nm (2), 6 ( 5) 表 4 水稻区提取分类精度评价 /% /% % Kappa (Continuum Removal),, (, 2005), (BNC), : BNC = (1 ( R/ R )) /(1 ( R / R )) (4), R, R, R c, R c (BNA), : BNA = (1 ( R / R')) / A (5), A c c 5 表 5 入选氮素和叶绿素吸收特征位置及波段范围 /nm /nm , , , 2060, 2130, , 2300,

13 1118 Journal of Remote Sensing 遥感学报 2009, 13(6) Curran(1989) : nm, nm,, 4.2, 6 6, 1,, , 3, 7, 670nm, 670Area(BNC), 3, 8 表 6 选用的植被指数 NDVI560_670 Normalized difference R560 R670 NDVI560 _ 670 = vegetation index R560 + R670 Rouse,1974 RVI560_670 Ratio vegetation index R560 RVI560 _ 670 = R670 Pearson&Miller, 1972 NRI Nitrogen reflectance R570 R670 NRI = index R570 + R670 Schleiche, 2001 CARI a R670 + b R700 Chlorophyll absorption CARI = 2 a + 1 R ratio index 670 a= ( R700 R550) /150, b= R a Kim, 1994 VARI_green Visible atmospherically R560 R670 VARI_green = resistant index R560 + R670 + R450 Anatoly, 2002 VARI_700 Visible atmospherically R R R450 VARI _ 700 = resistant index R R 1.3 R Anatoly, 表 7 与氮含量成显著相关的光谱特征参数及其复相关系数 (R 2 )(2005 ) PSSRb( ) PSNDb( ) CARI( ) NDVI560_670( ) RVI560_670( ) NRI( ) CARI( ) VARI_green( ) VARI_700( ) Area(BNC) Depth(BNA) , 660Depth(BNA)

14 : Hyperion 1119 表 8 与叶绿素含量成显著相关的光谱特征参数及其复相关系数 (R 2 )(2005 ) PSNDb( ) CARI( ) NDVI560_670( ) RVI560_670( ) NRI( ) CARI( ) VARI_green( ) VARI_700( ) Depth(BNA) , 660Area(BNC) , 660Depth(BNA) , 560nm 670nm NDVI, NDVI560_670( ), 5.2 Hyperion Hyperion, 8, (6) (7) 670Area(BNC) (R 2 ) 0.79, NDVI560_670( ) R , ( 6 7) N = Area(BNC) (6) CHL= NDVI560_670( ) (7) 7 NDVI560_670( ) Hyperion, ( 8 9), 0.06% 4.46%, ,,,, Area(BNC),, Hyperion,

15 1120 Journal of Remote Sensing 遥感学报 2009, 13(6) 8 670nm (R 2 ) 0.79, (2), 560nm 670nm, NDVI560_670, 3, 2,, NDVI560_670 R ,,,, REFERENCES 9, Hyperion,, : (1), 670nm 3, 2,,, Anatoly A, Gitelson Y, Yoram J K, Robert S and Don R Novel algorithms for remote estimation of vegetation fraction. Remote Sens. Environ, 80: Curran P J Remote sensing of foliar chemistry. Remote Sensing of Environment, 30, Datt B, McVicar T R, Van Niel T G, Jupp D L B and Pearlman J S Pre-processing EO-1 hyperion hyperspectral data to support the application of agricultural indices. IEEE Transactions on Geoscience and Remote Sensing, 41(6): FLAASH Module User s Guide, ENVI FL ASH Version 4.2, August, 2005 Edtion < < Huete A R, Liu H Q, Batchily K and VanLeeuwen W A comparison of vegetation indices global set of TM images for EOSMODIS. Remote Sensing of Environment, 59: Huete A R, Didan K, Miura T, Rodriguez E P, Gao X and Ferreira L G Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sensing of Environment, 83: Kim M S The Use of Narrow Spectral Bands for Improving Remote Sensing Estimation of Fractionally Absorbed Photosynthetically Active Radiation (fapar). Masters Thesis, Department of Geography, University of Maryland, College Park Li Y M, Ni S X and Wang X Z The robustness of linear regression model in rice leaf chlorophyll concentration prediction. Journal of Remote Sensing, 9(5): Liu L Y Applications of Hyperspectral Remote Sensing In Precision Agriculture. Postdoctoral research work report. Dis-

16 : Hyperion 1121 sertation of Institute of Remote Sensing Application Pearson R L and Miller L D Remote mapping of standing crop biomass for estimation of the productivity of productivity of the short-grass prairie. Pawnee National Grasslands, Colorado. Proceedings of the 8 th International Symposium on Remote Sens. Environ., ERIM International. Pu R L and Gong P Hyperspectal Remote Sensing and The Application. High Education Press, Beijing. Richard B. 2003, EO-1 User Guide v. 2. 3, and Rouse J W, Haas R H, Schell J A, Deering D W and Harlan J C Monitoring the Vernal Advancement of Retrogradation of Natural Vegetation. NASA/GSFC, Type III, Final Report, Greenbelt, MD, USA Schleicher T D, Bausch W C, Delgado J A and Ayers P D Evaluation and refinement of the nitrogen reflectance index (NRI) for site-specific fertilizer management ASAE Annual International Meeting. St. Joseph, MI, USA Shi R H, Niu Z and Zhuang D F Research on the effects of leaf biochemical concentrations on leaf spectra: case study of inversion of C:N ratio based on the absorption features centered at 2100nm. Journal of Remote Sensing, 9(1): 1 7 Zhu S P, Chen J, Chen Y, He P X and Li Y W The prototype research of precision agriculture computer integrated manufacturing system. Transactions of the CSAE, 16(4): 附中文参考文献,, , 9(5): , :,, nm., 9(1): 1 7,,,, , 16(4):